Safety production risk assessment method based on NLP big data

Abstract

The invention discloses a safety production risk assessment method based on NLP big data, and the method comprises the steps: 1, cleaning a corpus according to paragraphs, and removing numbers and special characters; 2, performing word segmentation according to a predefined word segmentation algorithm, and calculating an input vector V-input for each word by taking a paragraph as a unit; 3, obtaining hidden layer neuron values by adopting an artificial neural network; 4, calculating the neural network output of each word; 5, calculating an output probability value of each word through SoftMax;and 6, comparing output numerical values, and if not, adjusting the learning weight through a cost function and a gradient function; 7, repeating the first step to the sixth step to complete training; finally, according to the feature word vector distance and the word frequency, obtaining the weight value of the risk feature word as the input of risk assessment. According to the method, the associated feature words of the risk are obtained through the description data of the known accident, the related risk factor or risk source is obtained, the safety production risk data source input is increased, and the safety production risk assessment reliability is improved.

Description

technical field [0001] The invention relates to the technical field of NLP big data risk assessment, in particular to a safety production risk assessment method based on NLP big data. Background technique [0002] At present, the safety production risk assessment usually adopts risk assessment methods such as AHP (Analytic Hierarchy Process), HAZOP (Hazard and Operability analysis, hazard and operability analysis) and SCL. The AHP hierarchical analysis method is that the safety production risk is composed of different factors, and according to the interrelated influence and affiliation of the factors, the factors are aggregated and combined at different levels to form a multi-level analysis structure model, and from the lowest level ( For decision-making programs, measures, etc.) to the relative importance of the highest level (total goal) to assign different weights. This weight assignment usually adopts the method of expert experience value. The HAZOP risk assessment met...

AI Technical Summary

Problems solved by technology

The second is how to use NLP big data technology to solve the problem that low-frequency risk feature words are difficult to identify in some scenarios where accident data is insufficient (because some safety production accident scenarios 5 cannot be repeated)

Images